可视语音合成中口形特征点定位研究

Study of lip feature point location in visual speech synthesis

为了有效解决描述嘴唇轮廓的口形特征点定位问题,根据颜色空间中唇部和皮肤颜色的差异提出了一种新的从视频中定位口形特征点的方法。首先将视频文件分解为一帧帧的图片序列,定位并提取出图片中的口形区域,然后利用支持向量机(SVM)将口形区域分为唇部和非唇部区域,最后根据口形图像几何特征的先验知识实现口形特征点的定位。实验结果表明提出的方法具有较高的准确性和鲁棒性。

In order to solve the feature point location problem of describing the mouth appearance,a new method to locate lip feature point from video is proposed based on the fact that the lip color and skin-color are varied in the color space.This paper first extracts frames from digital video,locates and extracts oral area,then classifies oral area into lip and non-lip area by the Support Vector Machine.At last,based on knowledge about the structure of the mouth area to realize feature points location.The experiment results prove that the proposed method is accurate and robust.